agentic-scrum

🚧WIP🚧 This project is under active development. Expect breaking changes.

🤖 Agentic Scrum

Scrum optimized for AI agents for stable autonomous software development.

🚀 Getting Started

Optional dependencies: deno

For Claude Code Users

1. Use following prompts in Claude Code to add and install the plugin:

   /plugin marketplace add https://github.com/atusy/agentic-scrum
   

   /plugin install agentic-scrum@agentic-scrum
   

2. Run /scrum:init in your project to create scrum.ts

3. Prompt Claude Code to add a TODO in scrum.ts (i.e., PBI)

4. Run /scrum:go to start autonomous development

💡 Why Agentic Scrum?

Stable autonomous software development requires a structured process:

• 🧩 Incremental decomposition — Break problems down vertically (end-to-end slices) then horizontally (layers) for reliable resolution
• 🔍 Continuous quality inspection — Verify completed functionality meets standards
• 🔄 Adaptive prompting — Evolve instructions based on what works

Scrum provides exactly this structure, and AI agents understand it well.

Why adapt Scrum? Traditional Scrum assumes human limitations: time-boxed sprints, sprint point estimation, and synchronous ceremonies. AI agents don't have these constraints.

Agentic Scrum adapts the framework:

Traditional Scrum	Agentic Scrum
📅 Sprint = 2-4 weeks	⚡ Sprint = 1 PBI (any duration)
📊 Velocity planning	🚫 No estimation needed
👥 Team ceremonies	🤖 Autonomous coordination
📋 Sprint backlog items	🎯 Single focused goal

The result: continuous autonomous iteration with all the benefits of Scrum's inspect-and-adapt loop.

🏗️ Core Concepts

📄 Single Source of Truth: scrum.ts

All Scrum artifacts live in one TypeScript file that AI agents read and write:

const scrum: ScrumDashboard = {
  product_goal: { statement: "...", success_metrics: [...] },
  product_backlog: [...],      // Ordered by priority
  sprint: { goal: "...", subtasks: [...] },
  definition_of_done: { checks: [...] },
  completed: [...],            // Sprint history
  retrospectives: [...]        // Process improvements
};

🎭 AI Agent Roles

┌─────────────────────────────────────────────────────────────┐
│                  🧭 SCRUM MASTER                            │
│           Facilitates • Enforces • Removes Impediments      │
└─────────────────────────────────────────────────────────────┘
        │                                       │
        ▼                                       ▼
┌───────────────────┐                 ┌───────────────────────┐
│  📋 PRODUCT OWNER │                 │    💻 DEVELOPER       │
│                   │   ready PBI     │                       │
│  • Product Goal   │ ───────────────▶│  • TDD Cycle          │
│  • Backlog Order  │                 │  • RED → GREEN →      │
│  • Acceptance     │ ◀───────────────│    REFACTOR           │
│                   │   done PBI      │                       │
└───────────────────┘                 └───────────────────────┘

🔄 Status Lifecycles

PBI:      draft → refining → ready → done
Sprint:   planning → in_progress → review → done
Subtask:  pending → 🔴 red → 🟢 green → 🔧 refactoring → ✅ completed
                     │        │              │
                  (test)  (commit)      (commit×N)

📌 Key Principles

• ⚡ 1 Sprint = 1 PBI — Maximize iteration speed
• 📊 Order = Priority — Array position determines importance
• 🗂️ Git is History — No timestamps in dashboard
• 🔀 Behavioral ↔ Structural — Separate commits for features vs refactoring
• ✅ Commits at GREEN only — Never commit failing tests

📜 License

MIT © 2025 atusy